Prognostic utility and characterization of cell-free DNA in patients with severe sepsis

Abstract

Introduction: Although sepsis is the leading cause of death in noncoronary critically ill patients, identification of patients at high risk of death remains a challenge. In this study, we examined the incremental usefulness of adding multiple biomarkers to clinical scoring systems for predicting intensive care unit (ICU) mortality in patients with severe sepsis.

Methods: This retrospective observational study used stored plasma samples obtained from 80 severe sepsis patients recruited at three tertiary hospital ICUs in Hamilton, Ontario, Canada. Clinical data and plasma samples were obtained at study inclusion for all 80 patients, and then daily for 1 week, and weekly thereafter for a subset of 50 patients. Plasma levels of cell-free DNA (cfDNA), interleukin 6 (IL-6), thrombin, and protein C were measured and compared with clinical characteristics, including the primary outcome of ICU mortality and morbidity measured with the Multiple Organ Dysfunction (MODS) score and Acute Physiology and Chronic Health Evaluation (APACHE) II scores.

Results: The level of cfDNA in plasma at study inclusion had better prognostic utility than did MODS or APACHE II scores, or the biomarkers measured. The area under the receiver operating characteristic (ROC) curves for cfDNA to predict ICU mortality is 0.97 (95% CI, 0.93 to 1.00) and to predict hospital mortality is 0.84 (95% CI, 0.75 to 0.94). We found that a cfDNA cutoff value of 2.35 ng/μl had a sensitivity of 87.9% and specificity of 93.5% for predicting ICU mortality. Sequential measurements of cfDNA suggested that ICU mortality may be predicted within 24 hours of study inclusion, and that the predictive power of cfDNA may be enhanced by combining it with protein C levels or MODS scores. DNA-sequence analyses and studies with Toll-like receptor 9 (TLR9) reporter cells suggests that the cfDNA from sepsis patients is host derived.

Conclusions: These studies suggest that cfDNA provides high prognostic accuracy in patients with severe sepsis. The serial data suggest that the combination of cfDNA with protein C and MODS scores may yield even stronger predictive power. Incorporation of cfDNA in sepsis risk-stratification systems may be valuable for clinical decision making or for inclusion into sepsis trials.

Figure 1

Receiver operating characteristic (ROC) curves for cfDNA (∙), MOD score (□), and APACHE II score (Δ) to predict ICU mortality in 80 patients with severe sepsis. By using a binomial logistic model, we generated ROC curves to determine the predictive power of baseline values of cfDNA, MODS score, and APACHEII score for ICU mortality. The area under the curve (AUC) for cfDNA is 0.97 (95% CI, 0.93 to 1.00), for APACHE II score, is 0.64 (95% CI, 0.52 to 0.77), and for MODS score, is 0.63 (95% CI, 0.50 to 0.75).

Figure 2

Temporal changes in levels of cfDNA (A), PC (B), and MOD score (C) in 50 patients with severe sepsis. Survivors are shown by white circles (o), and nonsurvivors are shown by black circles (●). The number of patients at each time point (for which cfDNA, PC, or MODS values are available) is indicated in each graph. Data are shown as the mean ± SEM. The mean levels of cfDNA and PC in healthy volunteers (n = 14) is shown by the arrows.

Figure 3

Agarose gel electrophoresis of cfDNA from severe sepsis patients. cfDNA was purified from 250 μl of plasma, as described in Materials and methods, and 16 μl was loaded per lane. Lanes 1 to 5, cfDNA from severe sepsis patients (survivors); lanes 6 and 7, cfDNA from severe sepsis patients (nonsurvivors); lane 8, 100-bp DNA ladder.